File size: 5,573 Bytes
e4ba804
 
 
 
 
 
d958e80
17ea45c
 
 
 
 
b7bd7bd
17ea45c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed03fc1
17ea45c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4a2e5e1
17ea45c
 
 
 
 
 
 
 
 
d958e80
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-4B-Thinking-2507
library_name: transformers
---


## Latest News

* [2026-01-12]🚀🚀🚀 We have open-sourced **AgentCPM-Explore**, an agent foundation model with only **4B parameters**, together with its **entire training and inference infrastructure**. AgentCPM-Explore has successfully entered **8 classic long-horizon agent benchmarks**, including **GAIA,HLE, and BrowserComp**. AgentCPM-Explore achieves **SOTA performance at the same parameter scale** and demonstrates its **accurate deep research capabilities**, effectively breaking the performance bottleneck for **on-device agents**.

## Overview

Key highlights of AgentCPM-Explore include:

- The **first full-parameter 4B agent model** to rank on **8 long-horizon and complex agent benchmarks**, including **GAIA, HLE, and BrowserComp**, in the on-device setting.

- Capable of **over 100 rounds of continuous environment interaction**, supporting **multi-source information cross-validation**, **dynamic search strategy adjustment**, and **real-time verification of up-to-date information**, enabling sustained deep exploration until task completion.

- **Fully open-sourced end-to-end**, including (1) **AgentRL**, a fully asynchronous reinforcement learning framework for agent training, (2) **AgentDock**, a unified management and scheduling platform for tool sandboxes, (3) **AgentToLeaP**, a one-click evaluation platform for agent tool-learning capabilities. These components collectively support **community collaboration and custom extensibility**.

We elaborate on the entire construction pipeline of AgentCPM-Explore on [GitHub](https://github.com/OpenBMB/AgentCPM).


## Experimental Results
<table>
  <thead>
    <tr>
      <th>Model</th>
      <th>GAIA (text-only)</th>
      <th>BrowseComp</th>
      <th>BrowseComp (ZH)</th>
      <th>HLE</th>
      <th>Frames</th>
      <th>WebWalker</th>
      <th>Seal-0</th>
      <th>Xbench-DeepSearch</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td colspan="9"><strong>Closed-Source Models</strong></td>
    </tr>
    <tr>
      <td>Claude-4.5-sonnet</td>
      <td>71.2%</td>
      <td>19.6%</td>
      <td>40.8%</td>
      <td>24.5%</td>
      <td>85.0%</td>
      <td>/</td>
      <td>53.4%</td>
      <td>66.0%</td>
    </tr>
    <tr>
      <td>Gemini Deep Research</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>26.9%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
    </tr>
    <tr>
      <td>DeepSeek-V3.2</td>
      <td>63.5%</td>
      <td>67.6%</td>
      <td>65.0%</td>
      <td>40.8%</td>
      <td>80.2%</td>
      <td>/</td>
      <td>38.5%</td>
      <td>71.0%</td>
    </tr>
    <tr>
      <td>MiniMax-M2</td>
      <td>75.7%</td>
      <td>44.0%</td>
      <td>48.5%</td>
      <td>31.8%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>72.0%</td>
    </tr>
    <tr>
      <td>OpenAI-GPT-5-high</td>
      <td>76.4%</td>
      <td>54.9%</td>
      <td>65.0%</td>
      <td>35.2%</td>
      <td>/</td>
      <td>/</td>
      <td>51.4%</td>
      <td>77.8%</td>
    </tr>
    <tr>
      <td>GLM-4.6</td>
      <td>71.9%</td>
      <td>45.1%</td>
      <td>49.5%</td>
      <td>30.4%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>70.0%</td>
    </tr>
    <tr>
      <td>Kimi-Researcher</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>26.9%</td>
      <td>78.8%</td>
      <td>/</td>
      <td>36.0%</td>
      <td>69.0%</td>
    </tr>
    <tr>
      <td>Seed-1.8</td>
      <td>87.4%</td>
      <td>67.6%</td>
      <td>81.3%</td>
      <td>40.9%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
    </tr>
    <tr>
      <td colspan="9"><strong>Open-Source Models</strong></td>
    </tr>
    <tr>
      <td>MiroThinker 8B</td>
      <td>66.4%</td>
      <td>31.1%</td>
      <td>40.2%</td>
      <td>21.5%</td>
      <td>80.6%</td>
      <td>60.6%</td>
      <td>40.4%</td>
      <td>60.6%</td>
    </tr>
    <tr>
      <td>Tongyi DeepResearch 30B</td>
      <td>70.9%</td>
      <td>43.4%</td>
      <td>46.7%</td>
      <td>32.9%</td>
      <td>90.6%</td>
      <td>72.2%</td>
      <td>/</td>
      <td>75.0%</td>
    </tr>
    <tr>
      <td>ASearcher QWQ 32B v2</td>
      <td>58.7%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>74.5%</td>
      <td>/</td>
      <td>/</td>
      <td>51.1%</td>
    </tr>
    <tr>
      <td>iterresearch-30B-A3B</td>
      <td>72.8%</td>
      <td>37.3%</td>
      <td>45.2%</td>
      <td>28.8%</td>
      <td>71.0%</td>
      <td>/</td>
      <td>39.6%</td>
      <td>/</td>
    </tr>
    <tr>
      <td>WebSailor-V2-30B-A3B (RL)</td>
      <td>74.1%</td>
      <td>35.3%</td>
      <td>44.1%</td>
      <td>30.6%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>73.7%</td>
    </tr>
    <tr>
      <td>WebLeaper-30B-A3B-RUC</td>
      <td>73.2%</td>
      <td>38.8%</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>/</td>
      <td>48.6%</td>
      <td>72.0%</td>
    </tr>
    <tr>
      <td>WebDancer (QWQ-32B)</td>
      <td>51.5%</td>
      <td>3.8%</td>
      <td>18.0%</td>
      <td>/</td>
      <td>/</td>
      <td>47.9%</td>
      <td>/</td>
      <td>38.3%</td>
    </tr>
    <tr>
      <td>⭐ <strong>AgentCPM-Explore 4B</strong></td>
      <td>63.9%</td>
      <td>25.0%</td>
      <td>29.0%</td>
      <td>19.1%</td>
      <td>82.7%</td>
      <td>68.1%</td>
      <td>40.0%</td>
      <td>70.0%</td>
    </tr>
  </tbody>
</table>